Ocean Data

Campinas, São Paulo | Leveraging AI/ML for plastic marine debris

Ocean Data

High-Level Project Summary

Have you ever gone for a dive in the coast of America and, instead of fishes, found a sea of garbage? And, curiously, from brands located in countries on the other side of the world? Well, that's most likely due to the gigantic mass of garbage disposed daily in the ocean by many companies throughout the world, which tend to travel around the globe by sea and pollute marine life. For this, we propose Ocean Data, an open source API that uses ocean data and Machine Learning to predict the plastic footprints in the ocean, allowing non-governmental organizations, scientists and brands to hold the polluters accountable and help researchers understand the damage the plastic left on its path.

Link to Project "Demo"

https://youtu.be/r6x_2U2WlL8

Link to Final Project

https://www.figma.com/proto/JHIFYbz1fd1ARDlH79UOye/NASA-Ocean-Data?node-id=2%3A2&scaling=min-zoom&page-id=0%3A1&starting-point-node-id=2%3A2&hide-ui=1

Detailed Project Description

WHAT INSPIRED THE PROJECT

A study released by Oceans Asia estimates that in 2020 alone, 1.5 billion disposable masks ended up in the world’s oceans. Every year, at least 8 million tons of plastic, 80% of all marine debris from surface waters to deep-sea sediments, end up in our oceans. While a plastic object floats throughout the sea, up to 1,000,000 seabirds and 100,000 marine mammals and sea turtles die after ingesting or being entangled by it. And when it washes onto beaches, plastic pollution devalues real estate and affects tourism leading to financial losses of $6-billion to $19-billion.

Many variables must be considered to make the garbage path prediction, such as wind speed and direction, ocean tides and current velocity, weight and density of the object, amongst unpredictable phenomena such as storms and encounters with marine life. Furthermore, it is very difficult to identify single objects from remote sensing imagery and even more to track the path an object has taken in the ocean.

HOW WE DEVELOPED THIS PROJECT

Machine Learning - Neural Network

Our team developed a Neural Network machine learning model in Python language and TensorFlow framework.

The model was trained on Google Colab environment using the Global Drifter Program data of ocean surface drifting buoys by the National Oceanic and Atmospheric Administration.

The predictions are accessible through an API using the FastAPI: with the object present GPS coordinates and the date in which the object got there the model predicts the likely path taken by the garbage to reach this point.

We also created a high-fidelity prototype through Figma, which aims to represent the website for the Ocean Data solution. The goal of the website is to make the User Experience for scientists, researchers, NGOs and brands (our target audience) more friendly and intuitive as they get familiar with the platform and learn to use the API.

We plan to show the final product as a website map that helps tracking the plastic garbage origin.

Moreover, the API could be used to power citizen science Apps such as “Debris Tracker” to track the path of the garbage to the shore and further improve the user's engagement with UN sustainable development goals.

The API can be accessed through: https://github.com/AllanKamimura/Ocean_Data

Target audience:

Ocean Data is a free data processing API that aims to promote knowledge, helping researchers, scientists and NGOs to continue studying marine debris and to easily track plastics’ route so that the devastating consequences ocean pollution brings can be relieved. But, of course, Ocean Data’s API is open to anyone who wants to explore

Future improvements

Integrate historical data about wind velocity, ocean currents and water temperature into the model
Integrate object physical properties into the model

Space Agency Data

HOW WE USED SPACE AGENCY DATA IN THIS PROJECT

The model is trained on data from the Global Drifter Program, where we have time series data of the GPS position of a single object drifting through the ocean. This experiment give us an extensive amount of historical data ranging from as early as 1988 to present days.

We use this data to build a semi empirical model from where we can infer the movement and the trajectory of a single object.

Hackathon Journey

Team members:

Adriana Nagata (Brazil, São Paulo),
Allan Henrique Kamimura (Brazil, São Paulo),
Denizar Silva (Brazil, São Paulo),
Julia Gontijo Lopes (Brazil, Minas Gerais).

Our team was formed just a little before the day of Hackathon. We were complete strangers whose abilities and knowledge complemented each other in the course of the project. Defining the idea took a while, since Artificial Intelligence is such a broad and complex subject. But happily we agreed on a great idea and our work together was easy and objective.

We also got great help from our mentors, on defining our idea and identifying what the actual problem we were trying to solve.

Now that the project is over, besides learning more about working as a group, even if we were complete strangers, we learned much more about IA tools, its usability and more about the social part of the project. Marine life is indeed in more danger than we realize, and we must do something about it!